Technical

Enrichissement de transactions

Bank transaction enrichment

A layer on top of categorisation: it adds the merchant logo, company profile, MCC, geolocation, a normalised merchant identifier, subscription identification, and more.

Definition

Transaction enrichment sits on top of categorisation: it adds contextual metadata from external databases to each operation.

In concrete terms, a raw line becomes usable thanks to the merchant logo, the normalised trade name, the address and geolocation, the MCC, the company profile (SIREN, sector), the detection of a recurring subscription, and even the carbon footprint.

Categorisation vs enrichment

Often confused, but distinct:

CategorisationEnrichment
OutputA category ("Groceries › Supermarket")Metadata (logo, geo, SIREN…)
ModelML / rulesExternal database lookup + ML
DatabasesHistory of tagged labelsPappers, OpenStreetMap, MCC, merchant database
UseBudget, aggregation by categoryRich UI, precise identification, analytics

In practice, the two go together: a modern AISP, PFM or BFM offers both.

Common types of enrichment

  • Merchant identity: CB AMZN MKTPL DUBLIN IE becomes "Amazon", with logo and URL.
  • Company profile: for a corporate counterparty (transfer, direct debit), adding the SIREN, legal form, NAF sector and address (sources: Pappers, INPI).
  • MCC: a 4-digit code (ISO 18245) for card transactions (5411 supermarket, 4121 taxi, 7011 hotel).
  • Geolocation: the latitude/longitude of the point of sale, useful for the spending map and geographic fraud detection.
  • Subscription detection: spotting that a monthly NETFLIX €13.49 is a subscription, to offer dedicated management.
  • Income detection: identifying recurring incoming flows (salary, benefits) for scoring or budget protection.
  • Carbon footprint: estimating the CO2 of each expense (Carbo, Greenly, Doconomy) from an emission factor per sector.

Why it has become critical

Without enrichment, a transaction stays a rough draft: the user does not recognise the label (and complains to their bank), the app can show neither a logo nor a profile, behavioural analysis is poor and scoring loses precision. An enriched label ("Carrefour Express, 23/04, 75011 Paris, Groceries › Supermarket") is infinitely more useful than a raw CB CARREFOUR EXP 23/04.

Subscription detection: the star use case

Identifying recurring payments has become a service in its own right (Truebill in the US, Bankin' Subscriptions, Bridge Subscriptions):

  • spotting the recurrence (monthly, annual, weekly);
  • grouping label variants under a single subscription;
  • detecting price increases (Netflix going from €13.49 to €17.99);
  • assisting cancellation (the Chatel law, direct integration with certain merchants).

Data sources

  • MCC + merchant name: provided by the networks (Visa, MC, CB).
  • Pappers, INPI, RCS: French company profiles.
  • OpenStreetMap, Google Places, Yelp: geolocated points of sale.
  • Logos: public databases (Logo.dev, Brandfetch) or proprietary ones.
  • Subscription lists: maintained manually and supplemented by ML.
  • Carbon emission factors: ADEME (Base Empreinte), Doconomy.

What enrichment is not

  • Not a re-categorisation: it complements the category, it does not replace it.
  • Not always perfect: a very ambiguous label stays poor. Target: 80 to 95% of card transactions, 30 to 60% of transfers.
  • Not regulated: it relies on public or proprietary data, without any specific licence.
  • Not harmless for privacy: geolocating every transaction raises GDPR questions, hence the need for clear consent and purpose.

In the PSD2 ecosystem

Like categorisation, enrichment is outside DSP2 in the strict sense, but it is the major value-add of modern AIS solutions. Without it, every app would have the same raw data — which justifies the high valuations of the specialists (Bud, Tink, Heron Data).

Concrete examples

  • Specialists: Bud (UK), Heron Data (UK, B2B), Tink (Sweden), Powens and Bridge (FR), Yodlee Envestnet and MX (US).
  • Bankin': each transaction is displayed with a logo, a category and sometimes geolocation, for immediate recognition.
  • Pennylane / Qonto: accounting enrichment — supplier recognition (SIREN), reconciliation with OCR-scanned invoices, suggested allocation. A major time saver for freelancers.
  • Greenly / Carbo / Helios: carbon enrichment, with each expense becoming estimated kilos of CO2.
  • Subscription detection: Bankin' identifies ~30 subscriptions per user, including 3 to 5 forgotten ones — i.e. €50 to €150 of potential savings per year.
  • Cost: a full enrichment (logo, geo, subscription, MCC, SIREN) costs €0.002 to €0.02 per transaction, often included in an AIS + categorisation + enrichment offer.
  • Evolution: real-time enrichment (at the moment of the transaction) for instant budget alerts — already in place at Lydia and Revolut on their own accounts.

Sources

Go deeper

From the blog

Longer reads to go beyond the definition.

April 7, 2026
technical

Technical architecture of a PSD2 API: what to know before writing the first line of code — Read more

A technical overview of PSD2 for CTOs, developers and technical PMs: API standards, security (mTLS, eIDAS certificates), OAuth2, SCA flows, consent management, implementation pitfalls.

April 6, 2026
regulation

Understanding PSD2: what it changes for your customers and your business — Read more

Everything an executive, a product manager or a business team needs to understand about PSD2. Players, consent, opportunities, without technical jargon.